کنترل یادگیری تکراری اعمال شده به فرآیندهای دسته ای: مرور کلی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|27110||2007||13 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Control Engineering Practice, Volume 15, Issue 10, October 2007, Pages 1306–1318
With the recent emphasis on batch processing by emerging industries like the microelectronics and biotechnology, the interest in batch process control has been renewed. This paper gives an overview of the iterative learning control (ILC) technique, which can be used to improve tracking control performance in batch processes. The fundamental concepts and review of the various ILC algorithms are presented, with a particular focus on a model-based algorithm called Q-ILC and an application involving a rapid thermal processing (RTP) system. The study indicates that one can solve a seemingly very difficult multivariable nonlinear tracking problem with relative ease by intelligently combining the ILC technique with basic process insights and standard system identification techniques. Some related techniques in the literature are brought forth with the hope of unifying them. We aslo suggest some remaining challenges.
Batch processes have historically lagged continuous processes in terms of development and deployment of advanced optimization and control tools. Whereas significant developments have occurred during the past few decades in the industrial practice of continuous process control (Morari and Lee, 1999 and Qin and Badgwell, 2003), the same has not been the case for batch processes, which have continued to rely on old techniques like ladder-logics and PID control. Part of this can be attributed to the comparatively lower production volume through batch processing. Another reason for this may be that batch processes present a set of challenges uncommon in continuous processes, including nonstationary operating recipes, the consequent exposure to process nonlinearity, and significant variations in the initial charge condition (Berber, 1996). These challenges are not easily met by the standard linear optimal control theories and tools, which are widely adopted for continuous industrial process control today. However, the role of batch processing is ever-increasing in today's diversified manufacturing environment. Besides the fine or specialty chemicals, new industries that have emerged from the VLSI technology, bio-technology, and material science are mostly batch-processing-oriented. In accordance with its increased importance, its operation support tools need to be upgraded. Such a shift in the trend has already started taking place, as evidenced by the extensive use of run-to-run control and multivariate monitoring in some of the new industries. However, much more can be done, even with existing technologies today. For example, iterative learning control (ILC), the topic of this paper, has not enjoyed a serious look by the practitioners thus far despite its vast potentials for improving tracking control performance in batch processes. This paper presents an overview of ILC in the context of trajectory tracking problems in batch processes. Although basic theories of ILC have been firmly laid out in the literature, it is not always straightforward to apply them to achieve success in practice. ILC is discussed in the context of a multiple point temperature tracking problem in a rapid thermal processing (RTP) system. By doing so, the objective is to bring forth the unique capabilities of ILC for batch process control and at the same time some of the subtle challenges one may face in applying the technique. Fortunately, such challenges are not insurmountable and the standard linear ILC technique can provide an excellent performance for what appears to be a very difficult nonlinear trajectory tracking problem. Some related techniques like repetitive control and two-dimensional control are pointed out, highlighting the similarities and differences. Finally, some open issues left for future research are introduced.
نتیجه گیری انگلیسی
Iterative learning control (ILC) has great potentials for improving tracking control in batch processes. Though initially developed as a heuristic method for improving trajectory tracking performance of robot manipulators, two decades of research has laid solid theoretical foundations and generated insights needed for successful use in general tracking problems in batch processes. In particular, model-based algorithms like Q-ILC can address complex multivariable constrained systems and can be designed for significant robustness to model errors. The potentials and subtle challenges were demonstrated by presenting a case study involving an experimental RTP system. As turned out, one can effectively solve what appeared to be a very difficult multivariable, nonlinear tracking problem by combining the model-based ILC technique with some sound engineering judgment and creativity.An important assumption behind all current ILC algorithms is that the run length is fixed and the reference trajectory remains same. In many industrial batch processes, however, this assumption is oftentimes violated. Even though each batch run may slightly different, the basic pattern of the trajectory, such as hold-ramp-hold may not change. The main question is how one can translate the error trajectory from previous batch runs into an error trajectory for a new run, which may have a different length and reference trajectory. Another important issue is more systematic accounting for model errors in the ILC design. In particular, when the model error can be described quantitatively such as polytopic bounds for the dynamic gain matrix, it is desirable to use such information directly in the design. Because the batch system can be viewed as a simple integrating system along the batch index, derivation of robust ILC algorithms using the usual formulation like min-max optimization may prove to be more tractable. Some initial ideas along this direction can be found in Lee et al. (2000). Finally, the use of a nonlinear model within the existing ILC algorithms is conceptually straightforward (as mentioned in the discussion of BBO in the previous section) but comes with the usual slate of difficulties, which should come as no surprise to those familiar with the issues in nonlinear MPC.